The predictive value of respiration parameters and the Glasgow Goma score for mortality of sepsis patients with acute respiratory distress syndrome: Insights from the MIMIC-IV database

Pak J Med Sci. 2025 Nov;41(11):3152-3163. doi: 10.12669/pjms.41.11.13201.

ABSTRACT

BACKGROUND & OBJECTIVE: Acute respiratory distress syndrome (ARDS) frequently complicates sepsis, leading to significant morbidity and mortality. This study aimed to identify factors that predict short-term (30-day) and long-term (one-year) mortality in ARDS patients and to develop robust predictive models.

METHODOLOGY: This retrospective study, conducted from August 2024 to October 2024 in XinHua Hospital Affiliated with Shanghai Jiao Tong University, used data from the MIMIC database-specifically MIMIC IV 2.2-to identify sepsis patients who were diagnosed with ARDS within 24 hours of ICU admission. Univariate logistic regression was used to explore associations between respiratory parameters and the Glasgow coma scale (GCS) score (low group ≤ 12). Mortality at 30 days and one-year post-ICU admission was used as outcome measures. The dataset was balanced via synthetic minority over-sampling technique and split into training (70%) and validation (30%) sets. Variable selection was performed via the best subset, least absolute shrinkage and selection operator (LASSO), random forest, and boruta methods. Predictive models were developed and validated via calibration and decision curve analyses.

RESULTS: The cohort included 3,158 patients (58% female). Significant differences in PaCO₂ levels were detected between 30-day survivors and non-survivors (p>0.05), but not detected at one year (p>0.05). More patients with low GCS scores died within one year (20.9%) than survivors (17.1%; p=0.01), but no such association was found for 30-day mortality (p>0.05). The predictive models perform well for predicating short-term and long-term mortality and had AUCs of 0.820 and 0.790, respectively.

CONCLUSIONS: GCS scores were significantly associated with one-year mortality but not with 30-day mortality or with respiratory-related parameters. The developed predictive models demonstrated good performance. These findings aid in the treatment of ARDS patients with sepsis.

PMID:41394364 | PMC:PMC12697021 | DOI:10.12669/pjms.41.11.13201